Architecture and algorithm for parallel execution of a join operation

Abstract

The relational database system ARES which authors have developed can manage semantic aspects of data to be retrieved, and has two types of relations, namely, conventional and semantic relations. The semantic relation is attached to a conventional relation in terms of a join operation every time a flexible interpretation of queries, accepting a certain amount of ambiguity in query conditions, is required. In order to overcome performance degradation caused by the additional join operation, the authors present an algorithm for parallel execution of a join operation assuming an architecture composed of one master unit and n slave units which are linked to each other. Here, relations are distributed to n slave units by means of hash. According to the static analysis of performances, it is expected that the execution efficiency of a join operation is improved by the extent of o(n) and o(n2) compared with the cases of a single processor using hash for restricting source-target pairs of the subrelations, and that not using hash, respectively. Further investigation, however, is needed in a dynamic environment.